4x faster LLM inference (Flash Attention guy's company)

(together.ai)

96 points | by alecco 6 hours ago

6 comments

  • wishawa 3 hours ago
    Inference is impressively fast. But what about quality? In the Kimi vendor verifier (https://github.com/MoonshotAI/K2-Vendor-Verifier/), Together has one of the highest tool call failure rates (>300 failures over the benchmark, compared to 0-2 for the official API, groq, SiliconFlow, and Infinigence).
    • rfoo 2 hours ago
      If you compare "schema validation error count" plus "Count of Finish Reason others" then SiliconFlow and Infinigence is in the same bucket too. Maybe their API layer detected incorrect tool call and set finish reason to something else?

      IMO this likely is what you get from running the model correctly as-is (i.e. using the same weight and activation dtype), so Together is not bad.

      Moonshot AI themselves and Groq likely uses some sampler tricks to eliminate schema validation errors.

      So really the only thing this shows is: Nebius, Chutes, AtlasCloud could be running something else (for example further quantized model). Or bugs.

      • wishawa 1 hour ago
        Fair point. If Moonshot is holding back the true weights or inference techniques that affect correctness, then providers including Together should call them out on that. I for one would stop using Kimi if that is the case.

        Anyway, Novita is doing significantly better on the vendor verifier chart than Together, so the low quality must be partially Together's fault at least.

  • LogicFailsMe 57 minutes ago
    No barrier to entry whatsoever? Backprop on the speculative decoding weights during inference to improve their accuracy on a per application basis?

    Cool hack though, kudos. Wonder if they can make Groq or Cerebras do the same thing?

  • Havoc 2 hours ago
    >a faster speculator (also known as the draft model) proposes multiple tokens ahead, and the target model verifies them in parallel in a single forward pass

    TIL. Bit of an aha moment - never understood till now how a big model can verify faster than it can generate

    • woadwarrior01 1 hour ago
      As with almost everything else in CS, it's a tradeoff. Pre-fill is compute bound, decoding is memory bandwidth bound. Speculative decoding works when the draft model is more often right that wrong, because most architectures have a lot more compute, compared to memory bandwidth.
  • ashvardanian 3 hours ago
    Will need some time to go through the details, but it’s increasingly rare to see teams consistently delivering meaningful improvements in the open. Impressive work!
  • andblac 2 hours ago
    At first glance, this reminds me of how branch prediction is utilized in CPUs to speedup execution. As I understand it, this development is like a form of soft branch prediction over language trajectories: a small model predicts what the main model will do, takes few steps ahead and then verifies the results (and this can be done in parallel). If it checks out, you just jump forward, it not you take miss but its rare. I find it funny how small-big ideas like this come up in different context again and again in history of our technological development. Of course ideas as always are cheap. The hard part is how to actually use them and cash in on them.
  • petesergeant 3 hours ago
    > Built on top of Together Turbo Speculator, ATLAS reaches up to 500 TPS on DeepSeek-V3.1 and up to 460 TPS on Kimi-K2 in a fully adapted scenario — 2.65x faster than standard decoding, outperforming even specialized hardware like Groq

    and yet, if you click on: https://openrouter.ai/moonshotai/kimi-k2-0905

    You'll see Groq averaging 1,086tps vs Together doing 59tps. Groq and Cerebras often feel like the only games in town. I'd love that to be different (because I'd like more models!), but nobody else is coming close right now.

    Comparing how quickly gpt-oss-120b runs gives a broader picture: https://openrouter.ai/openai/gpt-oss-120b -- Vertex (Google) and SambaNova do pretty good on it too, but still, the difference between a top provider and an also-ran is giant.

    God I love OpenRouter.

    • KronisLV 1 hour ago
      > I'd love that to be different (because I'd like more models!), but nobody else is coming close right now.

      I'm currently on the Cerebras Code subscription for like 50 USD a month because it more or less makes the rate limits I used to deal with other platforms disappear (without making me spend upwards of 100 USD paying per token): https://www.cerebras.ai/blog/introducing-cerebras-code

      At the same time, their Qwen Coder 480B model is fine but I still find myself going for Claude or GPT-5 or Gemini 2.5 Pro for more complex issues (or ones where I need good usage of Latvian language), at least for programming tasks it'd eventually be super cool if they could offer more models.

      Or have some sort of a partnership with Anthropic or whoever, because getting my questions answered at around 500-1500 TPS is really, really pleasant, especially for agentic use cases with code modifications, even if I still bump into the 128k context limits occasionally.

    • meander_water 2 hours ago
      Interesting, if you take a look at the median throughput chart [0], groq goes insane after 7th Oct. Wonder what happened.

      [0] https://openrouter.ai/moonshotai/kimi-k2-0905/performance

      • sigmar 36 minutes ago
        2x jump overnight. new LPU hardware? I checked the speed for groq's gpt-oss-120B, Llama4-maverick, and Llama4-scout; none of them had a noticeable change this month
      • awestroke 2 hours ago
        Heavy quantization
    • immortal3 2 hours ago
      There's another angle to this comparison. Groq and Cerebras use custom chips, but I'm not sure about Together. In this case, Together is sharing results based on the B200 GPU. Another important point is the accuracy of these speed-ups compared to the baseline model. It's known that such tricks reduce accuracy, but by how much? Kimi has already benchmarked several providers. https://x.com/Kimi_Moonshot/status/1976926483319763130
      • rfoo 2 hours ago
        > It's known that such tricks reduce accuracy

        AFAIU, speculative decoding (and this fancier version of spec. decoding) does not reduce accuracy.

        • martinald 1 hour ago
          No it shouldn't do. "All" you're doing is having a small model run the prompt and then have the large model "verify" it. When the large model diverges from the small one, you restart the process again.
        • Der_Einzige 28 minutes ago
          It’s quantization which is crippling accuracy…
      • jsheard 2 hours ago
        > Groq and Cerebras use custom chips

        Not just custom chips, but custom chips which derive much of their performance from enormous amounts of SRAM. There's no denying that approach is fast, but it's also incredibly expensive, and SRAM scaling has slowed to a crawl so it won't get much cheaper any time soon.

    • senko 3 hours ago
      > You'll see Groq averaging 1,086tps

      What I don't understand is, Groq reporting 200tps for the same model: https://console.groq.com/docs/model/moonshotai/kimi-k2-instr...

      OpenRouter numbers look fishy.

    • jbellis 3 hours ago
      groq is quantizing, even though it's not labeled as such on openrouter (super frustrating)
      • bn-l 2 hours ago
        Do you have a source for that? They are pretty close to the ref implementation on moonshot’s ranking
    • Havoc 3 hours ago
      >Groq and Cerebras often feel like the only games in town.

      SambaNova should be similar...they've got a similar specialized hardware approach

    • p1esk 3 hours ago
      Do these numbers compare performance at the same cost?
      • petesergeant 1 hour ago
        You can see the cost in the links, and the answer is “pretty much” for the consumer. The backend maths, no idea.